[flink] Validate database existence when resolving catalog functions in FlinkCatalog

[pbanakar]

Purpose

Linked issue: close #3901

FlinkCatalog#listFunctions, #functionExists, and #getFunction resolved the built-in RoaringBitmap SQL functions (introduced in FIP-37) purely by function name, ignoring the database component of the request entirely. As a result, a fully qualified reference to a built-in function resolved successfully even when its database did not exist, e.g.:

SELECT fluss_catalog.nonexistent_db.rb_build(ARRAY[1, 2]);

This broke the database-scoped semantics Flink catalogs are expected to guarantee, and made it impossible to rely on the database component of <catalog>.<database>.<function> identifying an existing database.

Brief change log

• FlinkCatalog.java:
  • listFunctions(String dbName) now throws DatabaseNotExistException if dbName does not exist, before returning the built-in bitmap function names.
  • functionExists(ObjectPath) now returns false if the object path's database does not exist, before checking the built-in function map.
  • getFunction(ObjectPath) now throws FunctionNotExistException if the object path's database does not exist, before resolving the function class.
  • All three reuse the existing databaseExists(String) (already backed by admin.databaseExists()), so no new dependency or RPC path was introduced.
  • Built-in functions still resolve unqualified from any existing database — only the database-existence gate is new; no scoping to a specific database was added.

Tests

• Verified locally:
  • ./mvnw test -pl fluss-flink/fluss-flink-common -Dtest="FlinkCatalogTest" — BUILD SUCCESS
  • ./mvnw verify -pl fluss-flink/fluss-flink-common -Dit.test="Flink118CatalogITCase,Flink119CatalogITCase,Flink120CatalogITCase,Flink22CatalogITCase,RbFunctionsCatalogITCase" — BUILD SUCCESS

API and Format

no new public API is introduced.

Documentation

No user-facing documentation changes needed

[wuchong]

Thanks @pbanakar , I left some minor comments.

[wuchong]

Masking a missing database as a missing function can be misleading to users. I believe we should explicitly inform the user that the database does not exist. We can throw a CatalogException while wrapping it with a clear "database not found" message.

[pbanakar]

Updated getFunction to throw CatalogException with an explicit "Database X does not exist in catalog Y" message when the database is not found,

[wuchong]

Same here, we should throw CatalogException with database not found message rather than telling users the function not exist.

[pbanakar]

Same, updated this and also the unit test assertions accordingly

[pbanakar]

Thanks @wuchong! Quick heads up, throwing CatalogException in functionExists broke FlinkProcedureITCase#testCallNonExistProcedure in CI. Turns out Flink internally uses functionExists() as a boolean probe during function/procedure resolution, so throwing there short-circuits the lookup chain before it reaches Calcite's validator users get "Database ... does not exist" instead of the expected "No match found for function signature ...".

Can we revert just functionExists back to returning false and keep getFunction throwing CatalogException as you suggested, Does that split sound right to you, or would you prefer a different approach for functionExists?

[wuchong]

Thank you @pbanakar , I think you are right. I'm fine to revert the changes on functionExists.

[pbanakar]

thanks jark!, reverted it and i'll ping here once again once the ci passes

[pbanakar]

Hi @wuchong , Hit the same issue with getFunction as functionExists CatalogException here also breaks Flink's internal probe chain, need to revert getFunction also, you suggestions valuable here! if you know of a cleaner way to handel this instead of reverting? , happy to dig further, Please let me know!